GNNPipe: Scaling Deep GNN Training with Pipelined Model Parallelism

TL;DR

GNNPipe introduces a layer-level model parallelism approach for distributed GNN training, significantly reducing communication overhead and training time while maintaining model accuracy, and supports hybrid strategies for large graphs.

Contribution

First to apply layer-level model parallelism to GNN training, enabling scalable distributed training with reduced communication and improved efficiency.

Findings

Reduces per-epoch training time by up to 2.45x

Decreases communication volume by up to 22.89x

Maintains comparable model accuracy and convergence speed

Abstract

Communication is a key bottleneck for distributed graph neural network (GNN) training. This paper proposes GNNPipe, a new approach that scales the distributed full-graph deep GNN training. Being the first to use layer-level model parallelism for GNN training, GNNPipe partitions GNN layers among GPUs, each device performs the computation for a disjoint subset of consecutive GNN layers on the whole graph. Compared to graph parallelism with each GPU handling a graph partition, GNNPipe reduces the communication volume by a factor of the number of GNN layers. GNNPipe overcomes the unique challenges for pipelined layer-level model parallelism on the whole graph by partitioning it into dependent chunks, allowing the use of historical vertex embeddings, and applying specific training techniques to ensure convergence. We also propose a hybrid approach by combining GNNPipe with graph parallelism to handle large graphs, achieve better computer resource utilization and ensure model convergence. We build a general GNN training system supporting all three parallelism setting. Extensive experiments show that our method reduces the per-epoch training time by up to 2.45x (on average 1.58x) and reduces the communication volume and overhead by up to 22.89x and 27.21x (on average 8.69x and 11.60x), respectively, while achieving a comparable level of model accuracy and convergence speed compared to graph parallelism.